Set up Jira and Confluence

Complete this setup once on the computer that runs your AI client. Then add the client configuration.

Install uv

MCP Atlassian runs through uvx, which comes with uv.

curl -LsSf https://astral.sh/uv/install.sh | sh

Open a new terminal after installation. Then download the pinned package and check its command-line options:

uvx mcp-atlassian==0.23.0 --help

This command prepares the package in uv's tool cache. It does not register an MCP server with your client.

Find the executable path for the client configuration:

command -v uvx

Use the full path that this command prints. The client examples show common installation locations; replace them if your installation uses a different path.

Create the credentials

UC San Diego uses different authentication methods for these two services:

Your Triton AI Developer API key does not authenticate either connection.

Jira

1. Sign in to ITS PRO.
2. Select your avatar, then Profile > Personal Access Tokens.
3. Select Create token. Give the token a name and choose an expiration date.
4. Copy the token into your password manager. You cannot view it again after closing the dialog.

The PAT uses your existing Jira permissions. See Atlassian's PAT instructions.

Confluence

1. Open Atlassian API tokens with the account you use for Confluence.
2. Select Create API token for the direct site-URL setup below.
3. Give the token a name, choose an expiration date, and create it.
4. Copy it into your password manager. Record the email address of the Atlassian account that created it.

The direct ucsdcollab.atlassian.net/wiki configuration uses an API token without scopes. Atlassian routes scoped tokens through a different API URL. Do not paste a scoped token into this configuration without changing the URL. See Atlassian's API-token instructions.

Save a private environment file

Keep the tokens in one file outside your repositories. Each client will pass this file to MCP Atlassian.

Create the directory and open the file:

mkdir -p "$HOME/.config/tritonai"
touch "$HOME/.config/tritonai/atlassian.env"
chmod 600 "$HOME/.config/tritonai/atlassian.env"
open -e "$HOME/.config/tritonai/atlassian.env"

Paste this content into the file. Replace the three credential placeholders with your values. On Windows, save as atlassian.env, not atlassian.env.txt.

JIRA_URL=https://its-pro.ucsd.edu
JIRA_PERSONAL_TOKEN=YOUR_JIRA_PAT

CONFLUENCE_URL=https://ucsdcollab.atlassian.net/wiki
CONFLUENCE_USERNAME=YOUR_ATLASSIAN_EMAIL
CONFLUENCE_API_TOKEN=YOUR_CONFLUENCE_API_TOKEN

READ_ONLY_MODE=true
ENABLED_TOOLS=jira_search,jira_get_issue,confluence_search,confluence_get_page

Do not reuse the Jira PAT as the Confluence token. Keep the file private and out of version control, chat messages, and shared screenshots.

The --env-file option loads these values at server startup. In version 0.23.0, file values override matching process environment variables. See the server's configuration loader.

For a single-service setup, omit the other service's URL and credentials. Keep only that service's tools in ENABLED_TOOLS.

Connect your client

Use the guide for Claude Code, Claude Desktop, Codex, OpenCode, or Cline.

Every example uses absolute paths. Replace YOUR_NAME with your local account name and use the executable path you found earlier. Do not use ~, $HOME, or %USERPROFILE% inside the JSON or TOML examples.

If a path contains spaces, keep it in one argument. Do not add extra shell quotes inside a JSON string.

Allow selected write tools

To enable issue and page updates, change these two lines in atlassian.env:

READ_ONLY_MODE=false
ENABLED_TOOLS=jira_search,jira_get_issue,jira_update_issue,confluence_search,confluence_get_page,confluence_update_page

Restart the MCP connection after saving. Keep the client's tool-approval prompts enabled and review the target issue or page before approving an update.

These tools still use your Jira and Confluence permissions. See the pinned server's Jira tools and Confluence tools for their request fields.

Windows and WSL

Use the environment where the MCP client process runs:

If your editor connects to a WSL or remote development environment, configure MCP where its extension or agent runs. A Windows path does not identify a file inside that environment.